Modelling National Area Harvested of Paddy Using GARCH Methods (Generalized Autoregressive Conditional Heteroscedastic) Model.

Abstract

Time series data of national area harvested of paddy has high volatility and non homogeneous variance. A time series data with non homogeneous variance at time is called time series data with conditional heteroscedasticity. Time series analysis methods that can be used to overcome heteroskedasticity are GARCH models. However, the data of area harvested of paddy contained in the possibility of asymmetric volatility. To overcome the influence of asymmetry, some asymmetry GARCH models can be used, such as: exponential asymmetric GARCH model (EGARCH), quadratic asymmetric GARCH model (QGARCH), T-GARCH model, and non-linear asymmetry GARCH model (NAGARCH). This study aims to model the national area harvested of paddy by incorporating elements of varians heterogeneity and the influence of asymmetry on its data using five types of symmetry, asymmetry, and non-linear GARCH models, and find the best models of those five types of GARCH models. Model for national area harvested of paddy are ARIMA(2,0,0)(1,1,0)12 ‒ GARCH(1,2) and ARIMA(2,0,0)(1,1,0)12 ‒ QGARCH(1,2). Based on the mean absolute percentage error (MAPE) value to twenty two periods ahead, ARIMA(2,0,0)(1,1,0)12 ‒ QGARCH(1,2) better than ARIMA(2,0,0)(1,1,0)12 ‒ GARCH(1,2) but MAPE values for both models is quite high because there is some predicted value deviates quite far from the actual value. However, the value of MAPE to twelve periods ahead is low, 16.88% for ARIMA(2,0,0)(1,1,0)12 ‒ QGARCH(1,2). Furthermore, based on the value of mean absolute deviation (MAD) and mean square error (MSE), ARIMA(2,0,0)(1,1,0)12 ‒ QGARCH(1,2) also seems to be the better model than ARIMA(2,0,0)(1,1,0)12 ‒ GARCH(1,2). Thus, it can be concluded that the quadratic GARCH model is a fit model of national area harvested of paddy with a fairly good prediction results.